AUTHORS
Donna Zhe Sian Eng, MD, Steven Chan Sr, MD/PhD2, Maria Perusini, MD, Msc3, Marta Davidson, PhD MD4, Guillaume Richard-Carpentier Sr, MD5, Aniket Bankar, MD, MSc, DM, MBBS6, Hassan Sibai, MD, Msc, BSc7, Dawn Maze Sr, MD8, Karen W.L. Yee Sr, MD FRCPC9, Aaron Schimmer, MD PhD10, Andre C. Schuh Sr, BSc, MD, FRCPC11, Anne Tierens, MD, PhD12, Jose-Mario Capo-Chichi, MD, PhD13, Dennis Dong Hwan Kim, MD, PhD14
AFFILIATIONS
1 Princess Margaret Cancer Centre, Toronto, Canada
2 Princess Margaret Cancer Centre, Toronto, Canada
3 Princess Margaret Cancer Centre, Toronto, Canada, Princess Margaret Cancer Centre, Toronto, Canada
4 Princess Margaret Cancer Centre, Toronto, Canada
5 Princess Margaret Cancer Centre, Toronto, Canada
6 Princess Margaret Cancer Centre, Toronto, Canada
7 Princess Margaret Cancer Centre, Toronto, Canada
8 Princess Margaret Cancer Centre, Toronto, Canada
9 Princess Margaret Cancer Centre, Toronto, Canada
10 Princess Margaret Cancer Centre, Toronto, Canada
11 Princess Margaret Cancer Centre, Toronto, Canada
12 Princess Margaret Cancer Centre, Toronto, Canada
13 Princess Margaret Cancer Centre, Toronto, Canada
14 Princess Margaret Cancer Centre, Toronto, Canada
ABSTRACT
Background: European LeukemiaNet (ELN) measurable residual disease (MRD) framework relies on static NPM1 qPCR thresholds (e.g., >200 copies/10⁴ ABL, or a defined log-rise) to predict impending relapse risk in NPM1-mutated (NPM1m) acute myeloid leukemia (AML). A fixed numerical cutoff may not reflect 100% considering that molecular relapse is a kinetic process: the same transcript level does not always deliver the same prognostic implications if their kinetics are different. We hypothesized that the doubling time (DT) of NPM1 qPCR reflects the growth kinetics of residual NPM1m clones and adds prognostic information independent of the static >200-copies threshold, enabling a combined level-plus-kinetics risk model.
Patients & Methods: We analyzed newly diagnosed NPM1m AML patients treated at Princess Margaret Cancer Centre who achieved first complete remission (CR1) and underwent serial NPM1 qPCR (copies/10⁴ ABL) MRD monitoring since 2021. In total, 121 patients underwent 531 NPM1 qPCR assessments. 78 patients among whom 15 relapsed, with ≥2 serial measurements were analyzable for DT. For each patient, peak NPM1 level was classified by the ELN threshold (>200 vs ≤200 copies), and DT was calculated as DT = ln(2)/K, K = [ln(b)-ln(a)]/t (a, b successive levels; t days), the method previously validated for BCR-ABL1 qPCR (BJH 2022). Candidate DT cutoffs were swept and scored by sensitivity, specificity, predictive values (PPV/NPV), accuracy, F1, and relapse-free survival (RFS) log-rank significance. RFS, defined as time from CR1 to relapse or death, was the primary endpoint, estimated by the Kaplan-Meier method and compared by log-rank test. Risk discrimination was compared by C-index; recursive partitioning (survival tree) and multivariable Cox regression identified independent predictors of RFS.
Results: Among the 78 patients (median age 57.0y; FLT3-ITD 32.1%; median follow-up 33.1 months), a peak level ≥200 copies stratified RFS well (2-year RFS 0% vs 87.9% for <200 copies; p<0.0001; C-index 0.69). Sweeping candidate DT cutoffs showed a stable plateau of performance from approximately 15 to 150 days rather than a single sharp optimum; at 15 days, PPV 54.5%, NPV 86.6%, accuracy 82.1%, specificity 92.1%, sensitivity 40.0%, p=1.1x10^-4, HR 6.36 (95% CI [2.18-18.55]), C-index 0.67. The short-DT group (<15 days) had a 2-year RFS of 45.5% versus 88.6% for those with DT ≥15 days or no measurable rise; median DT in the high-risk group was 4.7 days (range: 0.6-14.0). Among the high-risk group who relapsed, clinical relapse occurred 4-9 weeks after the DT measurement (median 29 days, n=5).
The resulting four-leaf tree showed: level ≥200 copies & DT <15 days (n=4), 2-year RFS 0%; level ≥200 copies & DT ≥15 days (n=1), relapsed day 85; level <200 copies & DT <15 days (n=7), 2-year RFS 71%; level <200 copies & DT ≥15 days or undetectable (n=66), 2-year RFS 90%. Univariate Cox regression confirmed DT <15 days as a significant predictor of RFS (HR 6.36 [2.18-18.55], p<0.001); however, multivariable Cox regression could not establish DT as an independent predictor once peak level was included (HR 1.46 [0.23-9.50], p=0.69), nor could peak level ≥200 copies be assigned a stable multivariable hazard ratio in the same model. A combined level-plus-DT risk score nonetheless improved discrimination over either variable alone (C-index 0.72 vs 0.69 for qPCR level alone and 0.67 for DT alone).
DT's value is further refined among the 73 patients who never crossed 200 copies during surveillance: those with DT <15 days (n=7) had a 2-year RFS of 71% versus 90% for the remainder (n=66). A three-tier model (very high risk, both level ≥200 copies and DT <15 days, n=4; intermediate risk, either alone, n=8; low risk, neither, n=66) produced a clear outcome gradient: 2-year RFS 0%, 62.5%, and 90.1%, respectively (p=6.2x10^-15).
Conclusion: A rapid DT of NPM1 qPCR transcript (<15 days) was significantly associated with inferior RFS on univariate analysis but did not remain an independent predictor after adjusting for peak NPM1 level. Nevertheless, when combined into a three-tier model, a discrimination of 2-year RFS was found. These findings support DT as a clinically useful tool to predict imminent risk of relapse in NPM1m AML, and provides a guide to implement pre-emptive therapy within approximately a month in the short DT group. This result warrants validation in a larger cohort to confirm its prognostic relevance.
DISCLOSURES
Donna Zhe Sian Eng, MD: No relevant conflicts of interest to declare.